Superhuman's AI Agentic Framework Accelerates Productivity and Engineering
TL;DR
- Superhuman's AI integration prioritizes user productivity by never adding latency, enabling features like auto-labeling, smart summaries, and proactive follow-up nudges.
- Agentic search in Superhuman goes beyond simple retrieval, intelligently expanding half-formed queries to find specific information, reducing search time from minutes to seconds.
- Superhuman employs a tool-specific agentic framework, breaking down complex tasks into discrete actions executed by specialized tools for greater efficiency and control.
- The company rigorously evaluates AI models across multiple dimensions, including agent end-off and deep semantic search, using specific canonical queries to ensure high quality.
- Superhuman's infrastructure philosophy centers on local-first caching and hybrid search, ensuring fast, offline-capable interactions by storing data and embeddings on the device.
- AI adoption at Superhuman has accelerated development, increasing PR throughput by approximately 50% and reducing the time engineers spend on discovery and understanding code.
- The company believes AI will amplify engineers with strong fundamentals, widening the gap between those with deep understanding and those who lack it.
Deep Dive
Superhuman Mail is transforming the inbox from a simple communication tool into a powerful AI agent, leveraging its unique position as a private, contextual data source to deliver proactive assistance. This shift is driven by a philosophy of accelerating user productivity without introducing latency, a goal achieved through a sophisticated agentic framework that intelligently orchestrates specialized tools. The implications are profound, suggesting a future where AI EAs, powered by deeply integrated data across email, docs, and calendars, will redefine personal and professional workflows, while simultaneously widening the gap between truly skilled engineers and those who rely on superficial understanding.
The core of Superhuman's AI strategy lies in augmenting user productivity through subtle yet impactful AI interventions. Initially, this involved features like auto-labeling emails and smart summaries, designed to reduce cognitive load and streamline information processing. A more advanced iteration includes AI-powered nudges for follow-ups and auto-drafting replies, effectively acting as a proactive assistant. The "Ask AI" feature, a deep semantic search over an entire email history, has become particularly powerful. This agentic search capability, which can expand half-formed queries and navigate vast email archives through pagination, addresses the user's need to retrieve specific information buried within years of communication. This framework relies on a suite of specialized tools, each designed for distinct actions like detecting availability or composing emails, orchestrated by an agent that plans and executes tasks. The emphasis on "agent laziness" means the AI is designed to complete tasks proactively, minimizing user intervention and maximizing efficiency.
The implications of this approach extend to the very nature of software development and engineering culture. Superhuman's rapid iteration and adoption of AI tools, facilitated by a "free-for-all" policy and swift security approvals, have demonstrably increased engineer throughput, with PRs per engineer per week rising significantly. This highlights a second-order effect: AI tools not only accelerate coding but also enhance code discovery and understanding, reducing the time needed to onboard engineers and grasp complex codebases. Furthermore, the acquisition by Grammarly, integrating Superhuman's inbox intelligence with Grammarly's contextual understanding across various communication platforms, promises a more holistic AI EA. This unified stack could enable truly context-aware assistance, moving beyond email to encompass documents, calendars, and contracts. The ongoing challenge of constructing robust knowledge graphs from subjective user data, however, remains a significant hurdle, suggesting that while Superhuman aims to be a consumer of such graphs, the development of this foundational layer may fall to specialized engine companies.
Ultimately, Superhuman's vision positions the inbox as the central hub for future AI-powered personal assistance, leveraging its rich, private data to offer proactive, context-aware support. This proactive stance, contrasted with the reactive nature of general-purpose chatbots like ChatGPT, creates a competitive advantage. The success of this strategy hinges on Superhuman's ability to maintain its high-quality, user-centric approach, even as AI capabilities evolve. The drive for faster, more capable AI, coupled with a focus on user experience and engineering fundamentals, suggests that AI will increasingly differentiate truly exceptional engineers and products from the rest. The call for engineers who prioritize user experience and embrace AI as a tool for accelerated productivity underscores a shift towards a more dynamic and efficient software development landscape.
Action Items
- Audit authentication flow: Check for three vulnerability classes (SQL injection, XSS, CSRF) across 10 endpoints.
- Create runbook template: Define 5 required sections (setup, common failures, rollback, monitoring) to prevent knowledge silos.
- Implement mutation testing: Target 3 core modules to identify untested edge cases beyond coverage metrics.
- Profile build pipeline: Identify 5 slowest steps and establish 10-minute CI target to maintain fast feedback.
Key Quotes
"We started to really think about it like in q1 like a bunch of like people using some stuff and everything we didn't have any data just anecdotal feedback and all of that the first thing we've done is like cut the the red tape like for folks free for all I would approve the budget like in one hour you can try anything you want and deal with the security team 24 hours turnaround to get things approved from a security standpoint because you don't want to do some right crazy things"
Loïc Houssier explains that Superhuman's initial approach to AI adoption was to remove barriers. This involved allowing engineers to freely experiment with any AI tools and establishing a rapid 24-hour security approval process. This strategy aimed to foster widespread exploration and adoption of AI within the engineering team.
"I mean my main like central I would say piece of software that I use in my productivity tool is recast Ah yeah I'm I mean I'm a mac user so I use recast for the people that don't know recast it's basically like a web address that's applied on mac and I don't need bookmarks in my browser anymore"
Loïc Houssier describes Recast as a key productivity tool that functions as a web address manager for Mac users. This allows him to bypass traditional browser bookmarks, suggesting a shift towards more integrated and streamlined application access.
"I think that ai will do only one thing it will separate faster the good engineers from the bad engineers if you're a good engineer and you're using ai well you will be an amazing engineer if you're a poor lazy engineer and you don't want to understand things that you're doing ai will make you even worse because you will have the feeling that you get it but you're not being you're going like behind the magic behind the curtain behind things and how they work"
Loïc Houssier posits that AI's primary impact on the software industry will be to accelerate the differentiation between skilled and less skilled engineers. He argues that AI will amplify the capabilities of competent engineers, while those who lack fundamental understanding will be exposed more readily.
"The typical chatbot because more and more of the use case we see and people using ai inside superhuman is to query your emails a good example as how would I say like people will receive like a bunch of substack like a bunch of newsletter and we say so make great sometimes like the content is me I probably have like I don't know 30 40 subscription because everyone has like something interesting to say at some point and everything now I don't read them I auto archive those and like every week on a Friday I just like ask ai which is the name of the feature I asked my email hey tell me about like the summary of the newsletter of all the substack that I received this week what should I pay attention to and then I can deep dive"
Loïc Houssier illustrates a practical application of AI in Superhuman: querying email content. He uses the example of summarizing newsletters and substacks, allowing users to efficiently digest information and identify key points without reading every email.
"The main problem is like nailing the new user experience I mean open ai now you can do stuff with emails they're trying to do some stuff there like all those chatbots they try to be like the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the
Resources
## External Resources
### Books
- **"The Compound Startup"** - Mentioned as a model for operating as a startup within a larger company.
### Articles & Papers
- **"OpenTrust joins Docusign Global Trust Network"** (Docusign) - Referenced in relation to the acquisition of OpenTrust by Docusign.
- **"Grammarly acquires AI-email client Superhuman"** (TechCrunch) - Referenced in relation to Superhuman's acquisition by Grammarly.
### People
- **Loïc Houssier** - CTO of Superhuman Mail, guest on the podcast.
- **Rahul** - CEO of Superhuman, mentioned for his engagement with product quality and user experience.
- **Alessio** - Founder of Recurral Labs, co-host of the podcast.
- **Swix** - Editor of Latent Space, co-host of the podcast.
- **Shishir** - Mentioned as the board and executive team at Grammarly/Superhuman.
### Organizations & Institutions
- **Superhuman Mail** - Company where Loïc Houssier is CTO, focus of the discussion.
- **Docusign** - Mentioned in relation to Loïc Houssier's past experience and the complexities of enterprise software.
- **Grammarly** - Acquired Superhuman, mentioned in relation to the current company structure.
- **OpenAI** - Competitor and provider of AI models, discussed in relation to Superhuman's AI strategy.
- **Anthropic** - Competitor and provider of AI models, discussed in relation to Superhuman's AI strategy.
- **Gemini** - AI model from Google, discussed in relation to Superhuman's AI strategy.
- **Baseten** - Mentioned as an inference company used for running models.
- **Y Combinator (YC)** - Mentioned in relation to companies developing on-device models.
- **Google Cloud Platform (GCP)** - Mentioned as Superhuman's cloud provider.
- **Microsoft** - Mentioned in relation to Outlook.
- **Apple** - Mentioned in relation to iOS development.
- **Meta** - Mentioned in relation to React Native and Realm DB.
- **MongoDB** - Mentioned in relation to Realm DB's acquisition.
- **Product Board** - Mentioned in relation to Loïc Houssier's past experience.
- **France's defense industry** - Mentioned in relation to Loïc Houssier's early career.
- **National Football League (NFL)** - Mentioned as an example of a domain with specific data needs.
### Websites & Online Resources
- **docusign.com** - URL provided in relation to the OpenTrust acquisition.
- **linkedin.com/in/houssier/** - Loïc Houssier's LinkedIn profile.
- **x.com/latentspacepod** - Latent Space podcast's X (Twitter) handle.
- **latent.space/** - Latent Space podcast's website.
- **Recurral Labs** - Mentioned as Alessio's company.
- **Obsidian** - Mentioned as a note-taking application.
- **Figma** - Mentioned as a design tool.
### Other Resources
- **Applied cryptography** - Mentioned in relation to Loïc Houssier's academic background.
- **Nuclear submarine workflows** - Mentioned in relation to Loïc Houssier's early career.
- **E-signature startup** - Mentioned in relation to Loïc Houssier's entrepreneurial experience.
- **AI agentic framework** - Discussed as Superhuman's approach to AI in email.
- **Local-first architecture** - Discussed as a core principle for Superhuman's product.
- **Vector search** - Mentioned as a backend technology for AI.
- **Knowledge graphs** - Discussed as a potential future data model for productivity.
- **PKI (Public Key Infrastructure)** - Mentioned in relation to OpenTrust's business lines.
- **SSL certificates** - Mentioned in relation to OpenTrust's business lines.
- **FedRAMP** - Mentioned in relation to enterprise compliance requirements.
- **IMAP (Internet Message Access Protocol)** - Mentioned as a protocol handled by email providers like Gmail and Outlook.
- **SMTP (Simple Mail Transfer Protocol)** - Mentioned in relation to email server functionality.
- **GCP credits** - Mentioned as an incentive for using Gemini.
- **Price-per-trillion-tokens** - Discussed as a relevant cost metric for large-scale AI usage.
- **Price-per-million-tokens** - Discussed as a cost metric.
- **Compound startup model** - Mentioned as a strategy for operating within Grammarly.
- **V0** - Mentioned as a prototyping tool for Next.js.
- **Go** - Mentioned as Superhuman's backend programming language.
- **Swift** - Mentioned as Superhuman's iOS development language.
- **Kotlin** - Mentioned as Superhuman's Android development language.
- **React Native** - Mentioned as a cross-platform development framework.
- **Assembly** - Mentioned as a low-level programming language.
- **C** - Mentioned as a programming language.
- **SQL** - Mentioned in relation to database queries.
- **PCI DSS** - Mentioned in relation to enterprise compliance requirements.
- **Regional compliance** - Discussed as a complexity in enterprise software.
- **On-prem appliances** - Discussed in relation to Docusign's offerings.
- **Earthquake mitigation** - Mentioned in relation to Docusign's physical appliance deployment.
- **Data residency** - Discussed as a factor in enterprise software deployment.
- **Verticalization** - Mentioned in relation to Docusign's market approach.
- **Agent laziness** - Discussed as a problem to overcome in AI agent design.
- **Canonical queries** - Mentioned as a method for evaluating AI model quality.
- **Date reasoning** - Discussed as a challenge for large language models.
- **Local storage** - Mentioned as a feature of browsers.
- **Extensions** - Mentioned as a feature of browsers.
- **Operating system** - Discussed in relation to browser functionality.
- **Device drivers** - Mentioned in relation to operating systems.
- **App store** - Mentioned in relation to application platforms.
- **Graph database** - Mentioned as a technology not yet tried by Superhuman.
- **Taxonomy** - Discussed in relation to knowledge graphs and entity classification.
- **Entities** - Discussed in relation to knowledge graphs and personalization.
- **Projects** - Mentioned as an example of an entity.
- **Tasks** - Mentioned as an example of an entity.
- **Initiatives** - Mentioned as an example of an entity.
- **Templates** - Mentioned in relation to Obsidian.
- **SAS product** - Mentioned in relation to inferring taxonomy from roadmaps.
- **Authentication** - Mentioned as a common feature in SAS products.
- **Authorization** - Mentioned as a common feature in SAS products.
- **Billing** - Mentioned as a common feature in SAS products.
- **User management** - Mentioned as a common feature in SAS products.
- **Dashboards** - Mentioned as a common feature in SAS products.
- **Features** - Mentioned in relation to company surface area.
- **Surface area** - Mentioned in relation to company features.
- **AGI (Artificial General Intelligence)** - Mentioned in relation to AI remembering everything.
- **Compute** - Discussed in relation to graph recomputation.
- **Vector database companies** - Mentioned as potential builders of knowledge graph engines.
- **CRM (Customer Relationship Management)** - Mentioned in relation to mining emails for population.
- **Technical moat** - Discussed as no longer being a sustainable competitive advantage.
- **User experience** - Highlighted as a key differentiator for companies.
- **Interface** - Mentioned in relation to beautiful products.
- **Latency** - Discussed as an impact on end users.
- **Patagonia** - Mentioned as a location for remote AI team members.
- **Canada** - Mentioned as a location for remote AI team members.
- **Bay Area** - Mentioned as a competitive market for AI talent.
- **PNL (Profit and Loss)** - Mentioned in relation to Superhuman's financial structure within Grammarly.
- **Bugs** - Discussed in relation to increased feature shipping.
- **Marketing** - Discussed in relation to supporting increased feature shipping.
- **Support** - Discussed in relation to learning new features.
- **Salaries** - Discussed in relation to engineer compensation.
- **Maslow pyramid** - Used as an analogy for engineer motivations.
- **Dopamine** - Mentioned as a source of satisfaction for engineers.
- **Attrition** - Discussed in relation to work environment.
- **Assembly language** - Mentioned as a low-level programming concept.
- **Memory** - Mentioned as a fundamental computer concept.
- **Swaps** - Mentioned as a fundamental computer concept.
- **Serverless** - Discussed in relation to server infrastructure.
- **Abstraction** - Mentioned as a concept in programming.
- **Fundamentals** - Emphasized as crucial for engineers.
- **Magic behind the curtain** - Used to describe a lack of understanding of underlying processes.